Pablo Daniel Agüero

Prosody Generation for Speech-to-Speech Translation

This paper deals with speech synthesis in the framework of speech-to-speech translation. Our current focus is to translate speeches or conversations between humans so that a third person can listen to them in its own language. In this framework the style is not written but spoken and the original speech includes a lot of non-linguistic information (as speaker emotion). In this work we propose the use of prosodic features in the original speech to produce prosody in the target language. Relevant features are found using an unsupervised clustering algorithm that finds, in a bilingual speech corpus, intonation clusters in the source speech which are relevant in the target speech. Preliminary results already show a significant improvement in the synthetic quality (from MOS=3.40 to MOS=3.65)

Database Pruning for Unsupervised Building of Text-To-Speech Voices

Unit selection speech synthesis techniques lead the speech synthesis state of the art. Automatic segmentation of databases is necessary in order to build new voices. They may contain errors and segmentation processes may introduce some more. Quality systems require a significant effort to find and correct these segmentation errors. Phonetic transcription is crucial and is one of the manually supervised tasks. The possibility to automatically remove incorrectly transcribed units from the inventory will help to make the process more automatic. Here we present a new technique based on speech recognition confidence measures that reaches to remove 90% of incorrectly transcribed units from a database. The cost for it is loosing only a 10% of correctly transcribed units

Simulation of a non-invasive glucometer based on a microwave resonator sensor

In this paper a simulation of a microwave resonator sensor for constructing a non-invasive blood glucose meter is presented. A relationship between changes of the dielectric permittivity of the blood and the frequency response of S parameters of the sensor is observed. This can lead to a measuring procedure in which the glucose level present has a correlation with the value of the frequency resonance of the sensor. The test bank consists of a planar spiral microwave resonator over which the individual under test places his/her finger. This modifies the initial frequency resonance of the resonator because of the change produced in the measuring procedure over the dielectric permittivity of the resonator. Simulations show a correlation between dielectric permittivity blood changes, and changes in the value of the frequency resonance, in the frequency response of S parameters of the resonator.

A development and implementation of a tinnitus treatment method

Tinnitus is a physiological phenomenon where a person listens sounds which have not been generated by any external source. Today, many people suffer this condition. Although, in very few cases therapeutic methods completely eliminate tinnitus, it is possible to apply a variety of techniques to improve the quality of life of people with this condition. One of the most used methods to treat tinnitus consists of masking the tinnitus using an external sound. The main goal of this work is to present the development of a tinnitus treatment method, which optimizes the synthesized sounds in order to improve the lifes quality of the user. Subjective tests and experimental results are used to analyze the performance of the method.

A study of JEMA for intonation modeling

In the literature many intonation models are trained using parameters extracted sentence-by-sentence on contours interpolated in the unvoiced segments. This may introduce a bias in the final parameters and a reduction of the generalization of the model due to the increased dispersion of them. Recently, we have proposed JEMA, a joint extraction and prediction approach for intonation modeling that avoids such assumption. The parameter extraction and model training are combined in a loop where i) the model is successively refined, and ii) the parameters are extracted using information provided by the model. In this papers we present experiments based on synthetic data to evaluate this approach in a controlled environment. Both, the results with synthetic data and with natural speech, show that the use of JEMA is clearly superior to the standard estimation approach. The parameters are correctly extracted using several degrees of missing data (0% to 80%) and Gaussian noise. In fact, the study shows that including JEMA in the training algorithm is even more relevant than the selection of a particular representation of the intonation contours, as Fujisaki, Bezier, Tilt, or others.

Noise generator for tinnitus treatment based on look-up tables

Treatment of tinnitus by means of masking sounds allows to obtain a significant improve of the quality of life of the individual that suffer that condition. In view of that, it is possible to develop noise synthesizers based on random number generators in digital signal processors (DSP), which are used in almost any digital hearing aid devices. DSP architecture have limitations to implement a pseudo random number generator, due to it, the noise statistics can be not as good as expectations. In this paper, a technique to generate additive white gaussian noise (AWGN) or other types of filtered noise using coefficients stored in program memory of the DSP is proposed. Also, an implementation of the technique is carried out on a dsPIC from Microchip®. Objective experiments and experimental measurements are performed to analyze the proposed technique.

Flexible pseudorandom number generator for tinnitus treatment implemented on a dsPIC

Treatment of tinnitus with masking sounds has reach a significant developed in recent years. It is mainly because it has been possible to implement noise sinthesizers based on random number generators in digital signal processors (DSP), which form a part of almost any digital hearing aid device. One limitation of these methods is that limitations of the DSP architecture prevent pseudo white noise of being generated conform to a real white noise statistics. In this paper, a technique to generate adittive white gaussian noise (AWGN) or noise with uniform distribution using coefficients stored in memory of the DSP program is proposed. An implementation of the technique is carried out on a dsPIC from Microchip and subjective experiments and experimental measurements are performed to validate the performance of the developed technique.

Implementation of a Noise Reduction Algorithm in a Hearing Aid Device Based on a dsPIC

Nowadays there are many people affected by hearing impairments. The cost of assistive listening devices leaves out of reach for many people a technological solution. In the search of an economical alternative, low-cost DSPs arises as a good choice. However, it could be difficult to decide which has the best cost - effectiveness trade off among all devices offered in the market. The main goal of this work, then, is to determine the memory requirements of three noise reduction methods. It is done in order to select the algorithm that uses fewer resources, which will be implemented on a hearing aid device based on a low-cost DSP. This assistive listening device through digital signal processing techniques can get the basic features of a high-end assistive listening device. Objective experiments are performed to analyze the performance of the implemented system..

IMPROVING THE PERFORMANCE OF ANTI-SPAM FILTERS USING OUT-OF-VOCABULARY STATISTICS

This paper presents a feature based on out-of-vocabulary word statistics that complements the information sources used in the decision by state-of-the-art spam filters. The experiments included freely available spam filters as reference, SpamAssassin, Bogofilter, SpamBayes and SpamProbe, as well as a Naive Bayes classifier. The results show that the decision based on the proposed feature improves the performance of all spam filters under study.